Real Time Face Detection and Tracking on

It is possible to classify the face detection algorithms according to three categories. Very simple algorithms (based on averages of pixel) allowing the detection of face lling up most of the image, complex algorithms (PCA, neural networks) allowing to detect small and big faces and nally intermediate algorithms (often based on contours and skin color detection) allowing the localization of medium faces. Until now, only the algorithms of the rst category operated in real time on PC, the algorithms of the second and third category being implemented on expensive w orkstations dedicated to image processing. We propose in this article an algorithm belonging to the second category operating in real time (25Hz) in tracking mode on a common PC and able in search mode to locate faces diierently oriented at 10 detections per seconds. This algorithm is based on the approximation of the outlines of the face by an ellipse. A new optimized version of the GFHT (Generalized Fuzzy Hough Transform) which w e p r e s e n t here is implemented to locate quickly ellipses of various sizes.